Programmatic Ad Waste: Where the Money Actually Goes
The standard 'programmatic supply chain transparency' study, repeated by industry bodies every few years, keeps producing the same uncomfortable headline: of every dollar an advertiser spends, somewhere between $0.40 and $0.60 reaches the publisher. The rest evaporates into intermediaries, fees, and outright loss. The question worth asking is what each of those layers is actually doing.
The dollar's journey, layer by layer
Take a hypothetical $1.00 spend through a typical demand-side platform (DSP):
| Layer | Typical take | What you get for it |
|---|---|---|
| Agency / trading desk fee | $0.10 – $0.20 | Campaign management, creative, optimization decisions |
| DSP fee | $0.10 – $0.15 | Bidding infrastructure, audience graph, reporting |
| Ad-verification fee | $0.01 – $0.03 | Viewability, fraud, brand-safety measurement |
| Data-management / DMP fee | $0.02 – $0.10 | Third-party audience segments (when used) |
| SSP fee | $0.10 – $0.20 | Publisher-side auction matching |
| 'Tech tax' — unattributed | $0.05 – $0.15 | The infamous unaccounted gap that ANA studies keep finding |
| Publisher | $0.40 – $0.60 | The actual content that gets the impression |
The 'unattributed' line is the one industry studies keep flagging. It is not stolen money — it's the cumulative cost of resold inventory, header bidding rebid loops, multiple SSPs touching the same impression, and reconciliation failures across measurement systems. It is also where the most plausible cost reductions live, since by definition no one is providing a service for it.
The two big categories of waste, beyond fees
1. Made-for-advertising inventory. Sites built specifically to harvest programmatic spend, with thin or AI-generated content, aggressive ad density, and traffic acquired through arbitrage. The ANA study cited above estimates 21% of programmatic spend lands here. Practical fix: maintain a tight whitelist or use one of the MFA-exclusion services that have emerged since 2023.
2. Frequency over-targeting. The average user sees 6–9 impressions of the same campaign across the open web. Beyond 3–4, the marginal lift per additional impression collapses, and beyond 7 it can go negative (annoyance and ad-blocker installation). Cross-platform frequency capping is harder than single-platform capping because it requires unified identity, which after the deprecation of third-party cookies is genuinely difficult. Practical fix: budget caps per audience cohort, not just per campaign.
What actually moves the published-fee number
Three structural choices have a measurable impact on how much of the dollar reaches the publisher:
- Direct deals or curated marketplaces. Cuts SSP and DSP path complexity. PMPs (private marketplaces) and direct guaranteed deals typically move publisher take-rate from 50% to 65–75%.
- Self-serve DSP with no agency layer. If you have the in-house capacity, the agency fee disappears. Realistic for advertisers above $100K/month spend; below that, the operational overhead exceeds the savings.
- Walled gardens (Google, Meta, Amazon, TikTok). Inventory and tech are bundled, no independent SSP layer, no separate verification. Take rates are higher per platform fee but the chain is shorter; effective publisher take is 70–85% within the walled garden's own properties.
The frequent recommendation to 'go programmatic' or 'go walled garden' is a non-answer. Each model has structural advantages for different goals, and a sophisticated mix typically uses programmatic for prospecting at scale (where the open web's reach matters), walled gardens for retention and bottom-of-funnel (where targeting is sharper), and direct deals for premium environments where context matters as much as audience.
The headline number to monitor
If you measure one thing on programmatic spend, measure your working media percentage: of every dollar you commit, what percentage actually paid for an impression a real user saw? In 2024, the industry median was 36%. Best-in-class operators run at 55–60%. Below 30%, something is structurally wrong — either the chain is too long, or fraud is uncontrolled, or the inventory mix has drifted into MFA territory. Either way, that's where attention should go before tweaking creative or bid strategies.